Dysarthric speech recognition is a highly challenging task. The articulatory motor control problems associated with neuro-motor conditions produces large mismatch against normal speech. In addition, such data is difficult to collect in large quantities. This paper presents an initial attempt at the Chinese University of Hong Kong to develop an automatic speech recognition (ASR) system for the Universal Access Speech (UASpeech) task. A range of deep neural network (DNN) acoustic models and their more advanced variants based on time delayed neural networks (TDNNs) and long short-term memory recurrent neural networks (LSTM-RNNs) were developed. Speaker adaptation by learning hidden unit contributions (LHUC) was used. A semi-supervised complementary auto-encoder system was further constructed to improve the bottleneck feature extraction. Two out-of-domain ASR systems separately trained on broadcast news and switchboard data were cross domain adapted to the UASpeech data and used in system combination. The final combined system gave an overall word accuracy of 69.4% on the 16-speaker test set.
Cite as: Yu, J., Xie, X., Liu, S., Hu, S., Lam, M.W.Y., Wu, X., Wong, K.H., Liu, X., Meng, H. (2018) Development of the CUHK Dysarthric Speech Recognition System for the UA Speech Corpus. Proc. Interspeech 2018, 2938-2942, doi: 10.21437/Interspeech.2018-1541
@inproceedings{yu18b_interspeech,
author={Jianwei Yu and Xurong Xie and Shansong Liu and Shoukang Hu and Max W. Y. Lam and Xixin Wu and Ka Ho Wong and Xunying Liu and Helen Meng},
title={{Development of the CUHK Dysarthric Speech Recognition System for the UA Speech Corpus}},
year=2018,
booktitle={Proc. Interspeech 2018},
pages={2938--2942},
doi={10.21437/Interspeech.2018-1541}
}